From 911ea0c586099b52af8ec408678b33bc8f6ad63d Mon Sep 17 00:00:00 2001
From: Yash Jain <yashjainjain1704@gmail.com>
Date: Mon, 27 Jul 2026 05:19:22 +0100
Subject: [PATCH] [Bugfix][Frontend] Parse VLLM_USE_MODELSCOPE via vllm.envs
 (#5428)

Signed-off-by: Yash Jain <yashjainjain1704@gmail.com>
---
 .../test_omni_snapshot_download.py            | 84 +++++++++++++++++++
 vllm_omni/entrypoints/omni_base.py            |  7 +-
 2 files changed, 90 insertions(+), 1 deletion(-)
 create mode 100644 tests/entrypoints/test_omni_snapshot_download.py

diff --git a/tests/entrypoints/test_omni_snapshot_download.py b/tests/entrypoints/test_omni_snapshot_download.py
new file mode 100644
index 000000000..412c0f6d9
--- /dev/null
+++ b/tests/entrypoints/test_omni_snapshot_download.py
@@ -0,0 +1,84 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+"""``omni_snapshot_download`` must honor vLLM's ``VLLM_USE_MODELSCOPE`` semantics.
+
+vLLM treats the flag as enabled only for the literal string ``"true"``
+(case-insensitive). Reading ``os.environ`` directly made every non-empty value
+truthy, so an explicit opt-out such as ``VLLM_USE_MODELSCOPE=0`` still took the
+ModelScope path.
+"""
+
+import sys
+import types
+
+import pytest
+from vllm import envs
+
+from vllm_omni.entrypoints import omni_base
+
+pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
+
+MODEL_ID = "Qwen/Qwen3-Omni-30B-A3B-Instruct"
+
+
+@pytest.fixture
+def download_backend(monkeypatch: pytest.MonkeyPatch):
+    """Run ``omni_snapshot_download`` and report which backend it selected.
+
+    ModelScope is not a vLLM-Omni dependency, so the ModelScope branch is stubbed
+    into ``sys.modules``; without the stub it would raise ``ModuleNotFoundError``
+    instead of being observable.
+    """
+    # vLLM caches env lookups once a service is initialized; make sure this test
+    # reads the values monkeypatch sets rather than a cached snapshot.
+    envs.disable_envs_cache()
+
+    picked: list[str] = []
+
+    snapshot_module = types.ModuleType("modelscope.hub.snapshot_download")
+    snapshot_module.snapshot_download = lambda model_id: picked.append("modelscope") or model_id
+    hub_module = types.ModuleType("modelscope.hub")
+    hub_module.snapshot_download = snapshot_module
+    root_module = types.ModuleType("modelscope")
+    root_module.hub = hub_module
+    for name, module in (
+        ("modelscope", root_module),
+        ("modelscope.hub", hub_module),
+        ("modelscope.hub.snapshot_download", snapshot_module),
+    ):
+        monkeypatch.setitem(sys.modules, name, module)
+
+    monkeypatch.setattr(
+        omni_base,
+        "download_weights_from_hf_specific",
+        lambda **_kwargs: picked.append("huggingface"),
+    )
+
+    def run() -> str:
+        picked.clear()
+        omni_base.omni_snapshot_download(MODEL_ID)
+        return picked[0] if picked else "none"
+
+    return run
+
+
+@pytest.mark.parametrize("value", ["0", "1", "False", "false", "no", "off"])
+def test_non_true_values_do_not_enable_modelscope(monkeypatch, download_backend, value):
+    monkeypatch.setenv("VLLM_USE_MODELSCOPE", value)
+
+    assert envs.VLLM_USE_MODELSCOPE is False
+    assert download_backend() == "huggingface"
+
+
+@pytest.mark.parametrize("value", ["true", "True", "TRUE"])
+def test_true_values_enable_modelscope(monkeypatch, download_backend, value):
+    monkeypatch.setenv("VLLM_USE_MODELSCOPE", value)
+
+    assert envs.VLLM_USE_MODELSCOPE is True
+    assert download_backend() == "modelscope"
+
+
+def test_unset_defaults_to_huggingface(monkeypatch, download_backend):
+    monkeypatch.delenv("VLLM_USE_MODELSCOPE", raising=False)
+
+    assert download_backend() == "huggingface"
diff --git a/vllm_omni/entrypoints/omni_base.py b/vllm_omni/entrypoints/omni_base.py
index 44db512b2..423aa65b9 100644
--- a/vllm_omni/entrypoints/omni_base.py
+++ b/vllm_omni/entrypoints/omni_base.py
@@ -7,6 +7,7 @@ from collections.abc import Sequence
 from typing import Any, Literal
 
 import huggingface_hub
+import vllm.envs as envs
 from vllm.logger import init_logger
 from vllm.v1.engine.exceptions import EngineDeadError, EngineGenerateError
 
@@ -63,7 +64,11 @@ def omni_snapshot_download(model_id: str) -> str:
 
     # TODO: this is just a workaround for quickly use modelscope, we should support
     # modelscope in weight loading feature instead of using `snapshot_download`
-    if os.environ.get("VLLM_USE_MODELSCOPE", False):
+    # Read through ``vllm.envs`` so the flag keeps vLLM's semantics (only
+    # "true", case-insensitive, enables ModelScope). Reading the raw variable
+    # here made every non-empty value truthy, so ``VLLM_USE_MODELSCOPE=0``
+    # took the ModelScope path while the rest of the stack stayed on HF.
+    if envs.VLLM_USE_MODELSCOPE:
         from modelscope.hub.snapshot_download import snapshot_download
 
         return snapshot_download(model_id)